bump bootstrap compiler to 1.90 beta
There were significantly less `cfg(bootstrap)` and `cfg(not(bootstrap))` this release. Presumably due to the fact that we change the bootstrap stage orderings to reduce the need for them and it was successful 🙏
bootstrap: refactor mingw dist and fix gnullvm
Fixes https://github.com/rust-lang/rust/issues/144533
The first two commits are NFC and only clean up the code, paving the way for the third commit. That said, I think they are worthwhile even without that fix - reusing the same function for two different outcomes was confusing.
The third commit is the fix for https://github.com/rust-lang/rust/issues/144533, but due to the cross-compilation dance it requires a workaround to find the DLL since that logic really was meant only for Windows builders. That workaround is short-lived and will be removed as soon as gnullvm bootstraps itself.
Address libunwind.a inconsistency issues in the bootstrap program
We noticed when building rustc multiple time in a roll, some files will not be consistent across the build despite the fact that they are built from same source under the same environment. This patch addresses the inconsistency issue we found on libunwind.a, by sorting the order of the files passed to the linker.
Consolidate staging for `rustc_private` tools
This PR continues bootstrap refactoring, this time by consolidating staging for `Mode::ToolRustc` tools. This refactoring was in the critical path of refactoring `test`/`dist`/`clippy`/`doc` steps, and getting rid of the rmeta/rlib sysroot copy, because tools are pervasive and they are being used for a lot of things in bootstrap.
The main idea is to explicitly model the fact that a stage N `Mode::ToolRustc` tool always works with two different compilers:
- Stage N-1 rustc (`build_compiler`) builds stage N rustc (`target_compiler`)
- Rlib artifacts from stage N rustc are copied to the sysroot of stage N-1 rustc
- Stage N-1 rustc builds the (stage N) tool itself, the tool links to the rlib artifacts of the stage N rustc
Before, the code often used `compiler`, which meant sometimes the build compiler, sometimes the target compiler, and sometimes neither (looking at you, `download-rustc`). This is especially annoying when you get to a situation where you have an install step that invokes a dist step that invokes a tool build step, where *some* compiler is being propagated through, without it being clear what does that compiler represent. This refactoring hopefully makes that clearer and more explicit. It also gets rid of a few `builder.ensure(Rustc(...))` calls within bootstrap, which is always nice.
`Rustdoc` needs to be handled a bit specially, because it acts as a compiler itself, I documented that in the changes.
It wasn't practical to do these refactorings in multiple PRs, so I did it all in one PR. The meat of the change is 9ee6d1c1ed112c3dcfb5684b33772b136df0dca3.
I tested manually that `x build rustdoc` and `x build miri` still works even with `download-rustc`, although I cannot promise any extra support for `download-rustc`, IMO we will just have to reimplement it from scratch in a different way.
As usually, I did some drive-by refactorings to bootstrap, trying to document and clarify things, add more step metadata and tests.
Since these changes broke Cargo, which was incorrectly using `Mode::ToolRustc`, I also changed cargo to `ToolTarget` in this PR.
Best reviewed commit-by-commit (note that I renamed `link_compiler` to `target_compiler`, in accordance to the rest of bootstrap, in the last commit).
r? `@jieyouxu`
try-job: x86_64-gnu-aux
try-job: x86_64-msvc-ext1
Testing ui-fulldeps in "stage 1" actually uses the stage 0 compiler, so that
test programs can link against stage 1 rustc crates.
Unfortunately, using the stage 0 compiler causes problems when compiletest
tries to obtain target information from the compiler, but the output format has
changed since the last bootstrap beta bump.
We can work around this by also providing compiletest with a stage 1 compiler,
and having it use that compiler to query for target information.
Properly pass path to staged `rustc` to `compiletest` self-tests
Otherwise, this can do weird things like use a global rustc, or try to use stage 0 rustc. This must be properly configured, because `compiletest` is intended to only support one compiler target spec JSON format (of the in-tree compiler).
Historically, this was probably done so before `bootstrap` was really its own thing, and `compiletest` had to be runnable as a much more "self-standing" tool.
Follow-up to rust-lang/rust#144675, as I didn't realize this until Zalathar pointed it out in [#t-infra/bootstrap > Building vs testing `compiletest` @ 💬](https://rust-lang.zulipchat.com/#narrow/channel/326414-t-infra.2Fbootstrap/topic/Building.20vs.20testing.20.60compiletest.60/near/532040838).
r? ````@Kobzol````
Otherwise, this can do weird things like use a global rustc, or try to
use stage 0 rustc. This must be properly configured, because
`compiletest` is intended to only support one compiler target spec JSON
format (of the in-tree compiler).
Simplify library dependencies on `compiler-builtins`
The three panic-related library crates need to have access to `core`, and `compiler-builtins` needs to be in the crate graph. Rather than specifying both dependencies, switch these crates to use `rustc-std-workspace-core` which already does this.
This means there is now a single place that the `compiler-builtins` dependency needs to get configured, for everything other than `alloc` and `std`.
The second commit removes `compiler-builtins` from `std` (more details in the message).
The three panic-related library crates need to have access to `core`,
and `compiler-builtins` needs to be in the crate graph. Rather than
specifying both dependencies, switch these crates to use
`rustc-std-workspace-core` which already does this.
This means there is now a single place that the `compiler-builtins`
dependency needs to get configured, for everything other than `alloc`
and `std`.
We noticed when building rustc multiple time in a roll, some files
will not be consistent across the build despite the fact that they
are built from same source under the same environment. This patch
addresses the inconsistency issue we found on libunwind.a by sorting
the order of the files passed to the linker.
Otherwise, `compiletest` would have to know e.g. how to parse two
different target spec, if target spec format was changed between beta
`rustc` and in-tree `rustc`.
bootstrap: enable tidy auto extra checks on tools profile
alternative to https://github.com/rust-lang/rust/pull/144461
this won't affect CI or any `./configure` based workflows, and will also not affect every rust contributor like that PR will. a slower rollout of this feature should reduce disruption if issues are discovered with it.
r? ``@Kobzol``
Add explicit download methods to download module in bootstrap
This PR attempts to decouple the default initialization of the config object from parse_inner. It moves specific download methods, previously used during the initial config setup, into standalone functions outside the config implementation.
r? ``@Kobzol``
bors